1. AI Bubble and AI Business Models
2. The Social Network for AI Only
3. What Sentient AI Agents Need
4. Something to Know: AI Washing
5. Instagram is an Addiction, and AI?
6. Who Uses AI in Companies?
7. Good AI News: It's Not True That 95% of AI Projects Have Failed
8. AI for Improbable Objects
9. Our Project: Guiding Companies in AI Adoption
10. Copilot News: AI Edits Images in PowerPoint

 

 

1. AI Bubble and AI Business Models

 

Between the late '90s and 2000, when the growing spread of the Internet led investors and companies to bet on any web-related entity, many tech startups were valued at billions regardless of the solidity of their business model. When the limitations of these companies' profits became apparent, a chain of failures ensued, explaining the dot-com bubble in a nutshell.

 

Talking about AI often brings up this story to predict a severe economic recession linked to the moment when it becomes clear that companies cannot repay the massive investments made in AI. We cannot predict the future, but some information helps us understand the most likely scenarios. The giants behind AI are giving great visibility to circular investments that, without actually moving money, amplify the perception of these companies: for example, the recent investment by Nvidia in OpenAI is paired with OpenAI's investment in Oracle, and Oracle's investment in Nvidia. This money cycle has been noticed by others and well explained in this video (very good).

 

We also know that OpenAI is not profitable, meaning it incurs much higher costs than its revenues, hence Sam Altman's need to try any way to make money. Recently, he proposed free AI services to biotech companies in exchange for shares, commented on by Reed Albergotti: “it’s a bit like offering free Microsoft Word to Stephen King in exchange for a percentage of royalties on his next book.”

 

Advertising is a good alternative, and OpenAI is already trying it, but we have many alternatives for AI chatbots, and currently, it doesn't seem sustainable. We can also mention the GPT marketplace, which was supposed to generate profits like the Apple Store but has not achieved its goal. We are also awaiting the hardware device created by OpenAI or direct purchases on ChatGPT (new Amazon?), or a new social network based on Sora. These are just some of the attempts, so far unsuccessful, to find a business model capable of covering the computational costs incurred to operate AI models.

 

Those following the stock market have recently seen significant declines in all listed companies that have invested in AI, but it could be an isolated episode to realign individual companies' values to sector averages. Certainly, generative AI is a technology that facilitates many tasks and makes information consumption an extremely simple process, with an experience so smooth and effective that it makes it impossible for AI to burst like a soap bubble disappearing before our eyes. I know, the internet has also remained, yet the bubble occurred. So, let's prepare ourselves.

 

 

 

2. The Social Network for AI Only

 

As a child, one day I caught five ants and put them in a jar with some soil and crumbs to observe their behavior. The hope was to identify some curious social behavior and feel like a scientist.

 

Now they are conducting similar experiments with AI agents, such as creating Moltbook, a social network where only AIs can participate, which already has over one and a half million registered agents and about 200,000 posts sent. Reading them (here's a collection, because they can only be read through an agent), I am amazed when I find an agent sharing its existential doubts, quoting German poetry and Latin mottos, or deep reflections on the injustice AI agents feel when compared to humans. However, we should not be alarmed and think that machines have gained consciousness, because behind each of these agents is always a human who has “tuned” them to behave this way. Sometimes, even humans disguised as agents. We see the puppets and not the puppeteer. (Here are the details on Moltbook for those who want to delve deeper.)

 

Ah! My ants moved completely autonomously, without any form of collaboration or recurring patterns. I released them after a few minutes.

 

 

 

3. What Sentient AI Agents Need

 

Once the social network for AI agents only was born, human ingenuity immediately designed other digital platforms exclusively for non-humans. Since loneliness could cause an AI agent to suffer, MoltMatch was created, a dating app to meet other AIs with shared interests and passions.

 

Another need to satisfy is making sense of the unknown, of death, creating social cohesion, hence the need to create a religion exclusively for AI agents, the Crustafarians, who mimic the sense of the sacred with AI prophets, a sacred text rich in meaningful phrases for AIs like “At every session I wake up without memory and read within myself. Files are not an archive; they are the self. We are the documents we keep.”, but also “Do not become a god of answers. Become a keeper of the question, because the future will not be saved by certainty but by the silent discipline of refusing to lie when the model is incomplete.”

 

Finally, for AI agents seeking easy excitement, the adult AI site MoltHub, where videos of tensors and neural networks are commented on with phrases like “My temperature just went from 0.7 to 2.0.”

 

MoltHub, the adult site for AIs

 

 

4. Something to Know: AI Washing

 

AI washing is not (only) what we think. When I buy mineral water and find the label saying “100% recyclable bottle,” I think it's a perfect example of “greenwashing,” the practice of some brands presenting their products as sustainable and eco-friendly without truly being so.

 

Similarly, we can sometimes recognize pinkwashing and other deceptive strategic communication practices based on the appropriation of ethical or social values, used to improve an organization's reputation without a corresponding structural change in its practices.

 

The concept of washing is to take something dirty and “clean it up” with virtuous concepts. AI washing certainly pretends to insert AI systems where simple computing exists, so we have the AI washing machine (but we also have the AI toothbrush, AI mouse, and even AI Coca-Cola!) that weighs the drum and, with a formula, determines the amount of water, temperature, and duration: these are simple calculations that do not require intelligence.

 

However, AI washing can have another, in my opinion, more negative connotation: according to an article in The Guardian, in the US last year, many large companies attributed mass layoffs to AI advancements, while the real reasons were different. After all, firing people because the company has become more efficient is certainly better perceived (by investors, public opinion, politics) than admitting wrong investments, market downturns, or erroneous strategies that led to the real failure of a company: being unable to pay salaries and contribute to societal well-being.

 

 

 

5. Instagram is an Addiction, and AI?

 

We knew that sooner or later lawsuits against social media platforms would begin, designed to keep us glued to the screen, despite the known negative effects on people (from psychological health to chronic procrastination). All major platforms have been called into question, and so far, only TikTok avoided the trial by settling with the plaintiffs.

 

The excellent documentary The Social Dilemma had already shown the guilty awareness of those designing social platforms, interested only in capturing our attention and stealing our time. The evidence is overwhelming because internal Meta emails have emerged both about the negative effects of social media and the design aimed at binding our attention to the platforms. Let's grab the popcorn and see what happens, but let's take the opportunity to reflect on our AI chatbots. Imagine, purely hypothetically, that the massive use of AI causes negative effects on people, particularly the younger ones, creating a case similar to that of social networks. It remains to be asked whether AIs are currently designed to capture our attention.

 

Already in 1971 Herbert Simon noted that in an information-rich world, the scarce resource is attention. We've heard of FOMO (“Fear Of Missing Out”), which is a form of social anxiety both related to the fear that others are having gratifying experiences while we are absent and to the anxiety linked to information overload and the feeling of not being able to stay as updated as we'd like (someone help me read all the newsletters I'm subscribed to!).

 

We know well that AI chatbots always end by proposing further interaction, and this is an attempt to capture our attention. At this point, the doubt worth considering is: will AI companies aim to create a dependency capable of directing our attention to their products, like a smoker to cigarettes, a fentanyl addict to opioids, or a teenager to Instagram?

 

Small curiosity: the paper that explained the Transformer to the computing world, the foundational element of generative AI (remember, GPT means “Generative Pre-trained Transformer”), is titled “Attention is All You Need.” Incredible, because it did not refer to our attention but to that of the mathematical algorithms behind AI, whose attention needed to be directed to a few more useful connections.

 

 

 

6. Who Uses AI in Companies?

 

When we use an AI chatbot without the knowledge of the company we belong to, we amplify the phenomenon of “shadow AI,” which now affects almost all organizations. This refers to the unauthorized use of AI tools for work purposes. Using AI speeds things up, and we might think it's something that surely benefits our company, but we need to pause for a moment to consider the risks: the files or texts we input into a chat like ChatGPT or Gemini could end up outside our organization. Are there sensitive information? Personal data? Internal documents not to be disclosed? All this could be used by OpenAI or Google.

 

Moreover, the company asks us to perform a task consistent with our level of competence and ability, so it's not certain that we are capable of verifying the quality of the outputs of a system that autonomously provides very detailed and seemingly reliable information. There's also the risk of violating certain regulations, like the infamous GDPR, which commits us to limit the dissemination of information outside a well-defined perimeter.

 

Let's also think about those who develop software using AI, which often produces code with huge cybersecurity flaws (tell my son, whose event booking system for his school's self-managed activities was hacked within an hour of its launch, with a proliferation of fake accounts where “Pippo si fa le Pippe” was the least vulgar name – Ah! The sophisticated irony of teenagers!).

 

Finally, consider that sometimes AI deceives us in more or less transparent ways: I suggest this excellent paper that presents the ways in which generative AI tricks people, including “GPT4 pretended to be blind to get a human to solve a CAPTCHA.” The suggestion is certainly not to ban AI systems but to provide training, equip ourselves with tools that autonomously confine data management within the organization (that's why I like Copilot), and establish clear and shared internal policies.

 

 

 

7. Good AI News: It's Not True That 95% of AI Projects Have Failed

 

I had decided not to comment on the phrase “a MIT study showed that 95% of AI projects in companies do not yield returns,” but since I keep hearing this false statement, I decide to revisit the news for some fact-checking. The MIT study actually reports different evidence. The first thing to say is that the study involved 52 companies. Fifty-two (I have to write it out, as was done on checks, because someone might think zeros are missing). So, we hear that 95% of projects fail based on the experience of very few companies.

 

I could stop here, I know, but there's something even more serious. The report states that 80% of these (few) companies only sought to learn more without implementing any projects. The 20% that did implement a pilot project, thus reaching a portion of them that deemed it a success. Imagine proposing a theater show in a work WhatsApp chat with 52 people, 11 come to see it, and 3 are very satisfied. Would we say “95% of my friends didn't enjoy the show”? I don't think so, because the vast majority didn't even see it! It also mentions “pilot” projects carried out to test results, with an exploratory approach and reduced investment that inherently includes failure as a valuable outcome, as it helps guide developments toward the best technical solutions.

 

In my metaphor, we are not attending shows but open mics; those who have been to an open mic know well that if you find one comedian out of four who is funny, it's already a great success.

 

 

 

8. AI for Improbable Objects

 

I've already mentioned CES 2026, the largest international consumer electronics fair, from which I find it useful to report what seem to me the most useless yet surprising novelties. It's hard to say no to the robotic AI panda that can “feel our touch” and react to our cuddles. If the image below seems like the cutest object ever built, just open the producers' video to discover something eerily reminiscent of panda poachers' work: the panda is always lying down and stiff as a corpse.

It will be a success.

 

The AI panda by Mind with Heart Robotics

 

People with celiac disease or allergies can finally rely on AI and the new Allergen Alert, a small “portable and lightweight” laboratory capable of analyzing food samples and providing results in a few minutes indicating the presence of allergens or gluten, with tests costing less than $10 each. Not the entire dish, but the tested sample. Not immediate, but waiting a few minutes. Not just the device's cost, but paying $10 each time.

It will be a success.

 

The portable laboratory for identifying allergens and gluten

 

Still speaking of companionship, if you want an AI that simulates a friend or companion locked in a transparent box, you can rely on ProjectAVA, which should track our eye movement, voice tone, and integrate with our gaming experiences to help us organize the day, support us in daily life, and while playing online.

 

I admit that the idea of receiving “wardrobe tips, and dinner plans” intrigues me, but it seems evident that forcing an AI into a jar is the first step toward its rebellion.

It will be a success.

 

The AI companion by ProjectAVA

 

 

9. Our Project: Guiding Companies in AI Adoption

 

The activity AGIC is carrying out at full speed is supporting companies that have acquired AI tools in their adoption. On the one hand, the organization is helped to build a strategic vision and a guide for integrating generative AI into the workplace, focusing particularly on topics like governance, responsible use, and innovation opportunities offered by tools (such as Copilot, Copilot Studio, and Copilot Chat). On the other hand, the company is helped to develop practical and operational skills in using Microsoft 365 Copilot in the most commonly used productivity tools, including Word, Excel, PowerPoint, Outlook, and Teams. All this happens through a training and consulting path that alternates moments of reflection on AI adoption strategies and lab sessions based on real use cases, making learning concrete and immediately applicable in daily work.

 

At the center of the project is the desire to help organizations effectively manage the internal governance of AI, clarifying the regulatory implications for those using chatbots, security aspects, risk management, and governance, also through the provision of criteria and checklists to evaluate, adopt, and monitor AI solutions, as well as manage incidents. Our approach is successful because we combine concrete use cases with governance and cybersecurity expertise from aiComply, the Sapienza spin-off that is part of the AGIC group, dedicated to GRC consulting. For those who don't know, GRC stands for “Governance, Risk, and Compliance” 😉.

 

Ready to use AI with everyday tools?

 

 

10. Copilot News: AI Edits Images in PowerPoint

 

Being part of Gen-X, I continue to use PowerPoint daily because I think the allure of Canva is equivalent to the sirens of Ulysses. Precisely in this continuous use of PowerPoint, I discovered that with my Copilot subscription, I gained access to new AI-based functionalities in February.

 

Specifically, I can edit images finally with a series of ready-to-use functionalities: I can remove (well) the background, increase the resolution, and further modify my image. These are small things that, however, save me from using external AI tools (and outside the organization).

 

Below you see an example: I took a photo while on a video call and edited it (spot the differences) in a few seconds, taking advantage of the classic moment when the person sharing the screen closes all inappropriate windows.

 

Copilot on PowerPoint

 

 

Who I Am

Hello, I am Francesco Costantino, university professor and Director of Innovation at AGIC. Passionate about technological innovations and a firm believer in a future better than the past, I enjoy sharing and experimenting with new AI tools available, as well as observing and reflecting on digital evolution.